The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash scripts/sync-shared-content.sh as part of its registration and validation phase. This is an expected action within the project's workflow to synchronize shared principles and validate the structure of newly forged skills.
[DATA_EXPOSURE]: The skill reads several project configuration files, including CLAUDE.md, plugin.json, and existing SKILL.md files. This is done to maintain project conventions and verify the environment. It explicitly includes security guidelines (Shared Principle #4) that forbid agents from including credentials or PII in prompts, messages, or artifacts.
[DYNAMIC_EXECUTION]: The skill generates new SKILL.md and persona files at runtime. This process is heavily audited by internal agents ('Interrogator' and 'Forge Auditor') which review draft prompts and structural compliance before files are written to disk. This multi-layered review serves as a defensive measure against generating unsafe or malformed agent instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (the concept prompt) to generate new agent instructions. It addresses this risk through the 'Interrogator' agent, which is specifically tasked with pressure-testing draft prompts against adversarial questions and failure modes (Iron Law #05) before finalizing the skill.

create-conclave-team